fix(catalog): add GLM-5.2 to ZAI provider

[LineZore]

What

GLM-5.2 was missing from the ZAI provider model list, so users authenticated with a ZAI API key could not see or select it.

Closes #542.

Root cause

The ZAI provider in src/tau_coding/data/catalog.toml had a hand-maintained models list that stopped at glm-5.1. GLM-5.2 (Z.ai coding plan) was never added, so it did not appear after /login.

Note: the catalog is static — there is no remote /models discovery — so new models must be added here explicitly.

Verified model specs (from Z.ai docs)

Cross-checked against Z.ai documentation, not copied from sibling entries:

• model id: glm-5.2 — confirmed to exist
• context window: 1,000,000 — Z.ai advertises GLM-5.2 as "supporting up to 1M context". This is a GLM-5.2-specific claim; the other GLM-5 models (e.g. glm-5.1) are not advertised as 1M, which is why they remain at 200000.
• max output tokens: 131072 (128K) — matches GLM-5.1
• reasoning: true (supports reasoning_effort)
• input: text only (the *-turbo vision variant is separate)
• cost: all zero — ZAI uses a coding-plan credit model, consistent with the existing GLM-5 entries
• compat: supportsDeveloperRole = false, thinkingFormat = "zai", zaiToolStream = true — same as glm-5.1 / glm-5-turbo (same protocol)

Changes

Three additions to the zai provider block in src/tau_coding/data/catalog.toml:

1. models list: add "glm-5.2" (between glm-5.1 and glm-5v-turbo)
2. context_windows: add "glm-5.2" = 1000000
3. new [providers.model_metadata."glm-5.2"] block

default_model is intentionally left as glm-5.1 — this PR only makes GLM-5.2 selectable; changing the default is a separate behavioral decision for maintainers.

Verification

• uv run pytest tests/test_provider_catalog.py tests/test_provider_config.py → 107 passed
• Runtime smoke test: catalog loads, glm-5.2 appears in zai.models, context_windows, and model_metadata with consistent values; default_model validation passes
• uv run ruff check → clean